TrackingMiM: Efficient Mamba-in-Mamba Serialization for Real-time UAV Object Tracking

2025-07-02

AI Summary

To address the high computational complexity of Vision Transformers (ViTs) and temporal inconsistency in existing Mamba-based approaches for real-time UAV target tracking, this paper proposes a nested Mamba-in-Mamba (MiM) architecture. MiM employs a spatiotemporally decoupled sequence scanning mechanism that separates image sequences into independent spatial and temporal modeling pathways; it further introduces a template frame as a persistent query to explicitly enforce temporal continuity in Mamba scanning—marking the first such explicit modeling of temporal coherence in tracking tasks. This design enables efficient cross-frame information synergy between template and search regions while preserving long-term consistency. Evaluated on five mainstream UAV tracking benchmarks, MiM achieves state-of-the-art accuracy with significantly improved inference speed, satisfying real-time operational requirements.

Technology Category

• Computer Vision: Motion & Tracking
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Web Mining and Content Analysis: Topic discovery and tracking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

The Vision Transformer (ViT) model has long struggled with the challenge of quadratic complexity, a limitation that becomes especially critical in unmanned aerial vehicle (UAV) tracking systems, where data must be processed in real time. In this study, we explore the recently proposed State-Space Model, Mamba, leveraging its computational efficiency and capability for long-sequence modeling to effectively process dense image sequences in tracking tasks. First, we highlight the issue of temporal inconsistency in existing Mamba-based methods, specifically the failure to account for temporal continuity in the Mamba scanning mechanism. Secondly, building upon this insight,we propose TrackingMiM, a Mamba-in-Mamba architecture, a minimal-computation burden model for handling image sequence of tracking problem. In our framework, the mamba scan is performed in a nested way while independently process temporal and spatial coherent patch tokens. While the template frame is encoded as query token and utilized for tracking in every scan. Extensive experiments conducted on five UAV tracking benchmarks confirm that the proposed TrackingMiM achieves state-of-the-art precision while offering noticeable higher speed in UAV tracking.

Problem

• Overcome quadratic complexity in Vision Transformers for real-time UAV tracking
• Address temporal inconsistency in Mamba-based tracking methods
• Develop efficient nested Mamba architecture for spatial-temporal coherence

Innovation

• Mamba-in-Mamba architecture for efficient tracking
• Nested Mamba scan for temporal-spatial coherence
• Template frame encoded as query token